Integrated twitter analysis to distinguish systems thinkers at various levels: a case study of COVID-19
Applied Network Science volume 8, Article number: 12 (2023)
Systems Thinking (ST) has become essential for practitioners and experts when dealing with turbulent and complex environments. Twitter medium harbors social capital including systems thinkers, however there are limited studies available in the extant literature that investigate how experts' systems thinking skills, if possible at all, can be revealed within Twitter analysis. This study aims to reveal systems thinking levels of experts from their Twitter accounts represented as a network. Unraveling of latent Twitter network clusters ensues the centrality analysis of their follower networks inferred in terms of systems thinking dimensions. COVID-19 emerges as a relevant case study to investigate the relationship between COVID-19 experts’ Twitter network and their systems thinking capabilities. A sample of 55 trusted expert Twitter accounts related to COVID-19 has been selected for the current study based on the lists from Forbes, Fortune, and Bustle. The Twitter network has been constructed based on the features extracted from their Twitter accounts. Community detection reveals three distinct groups of experts. In order to relate system thinking qualities to each group, systems thinking dimensions are matched with the follower network characteristics such as node-level metrics and centrality measures including degree, betweenness, closeness and Eigen centrality. Comparison of the 55 expert follower network characteristics elucidates three clusters with significant differences in centrality scores and node-level metrics. The clusters with a higher, medium, lower scores can be classified as Twitter accounts of Holistic thinkers, Middle thinkers, and Reductionist thinkers, respectfully. In conclusion, systems thinking capabilities are traced through unique network patterns in relation to the follower network characteristics associated with systems thinking dimensions.
Experts and practitioners need to address the increasing challenges of today’s sociotechnical systems while maintaining and elevating their performances under increasing complexities. These challenges include (1) requirement of high level systems integration to serve overarching goals, (2) lack of clarity to allow for decision support and commitment to alternative courses of action, (3) uncertainty caused by incomplete knowledge of systems and the unintended consequences, and (4) the effects of interdependence where there is mutual influence among systems and their related elements making analysis difficult (Boardman and Sauser 2006; Keating 2008; Nagahi et al. 2020a, b; Ullah Ibne Hossain et al. 2020). These four elements are likely to escalate as the interdisciplinary system problems of the twenty-first century continue to blur the lines between technical, social, managerial, and policy considerations (DeLaurentis and 2005, January. 2005; Gorod et al. 2008; Jaradat et al. 2017; Nagahi et al. 2020c; Karam et al. 2020). In response to these challenges, it is necessary to develop tools, pipelines and skills to support a more holistic, i.e. systemic, approach when dealing with complex system problems.
Sociotechnical failures disclose technical, sociopolitical, and power elements, or interactions between these elements (Nagahi et al. 2020a; Ullah Ibne Hossain et al. 2020; Frank 2006; Clegg 2000). A holistic frame of reference is necessary to provide effective solutions to address such rigorous failures. Consequently, studying the relationship between Systems Thinking (ST) skills and social media analytics, i.e. Twitter analysis, combines some social and technical aspects that might uncover information to support effective management of complex multidimensional systems. For example, the assessment of ST skills using social media analytics can help governments and the public distinguish the more accurate and systemic responses to the COVID-19 global pandemic to make informed decisions better in this critical and life-matter situation; given the contradictory statements, actions, and mismanagement of the pandemic at its early stage.
Although much has been written about systems thinking, the impact of systems thinking skills of COVID-19 experts on their tweets and Twitter follower networks to ensue is not elaborated.
This study aims to address the interdisciplinary literature gap in complex systems where ST skills are deciphered using social media analytics including Twitter and follower network analysis. Unraveling of latent Twitter network clusters is followed by the centrality analysis of the follower networks inferred in terms of systems thinking dimensions. A ST skills instrument (Jaradat 2015) is used to denote levels of systems thinking. The literature review identifies research gaps relevant to evaluating if experts' system thinking skills can be revealed through Twitter analysis. This article presents an overview of the relevant literature to integrate ST and social media analytics, followed by the data collection procedure, extraction of Twitter features for network construction and clustering, and follower network analysis. The results are analyzed and interpreted. Conclusions, limitations, and future work ensue.
Review of literature
Network representation encodes individuals as nodes, and relations between them as edges. The simple yet powerful representation encompasses information hidden under micro-to-macro scale network features. Microscale features emphasize certain node or edge characteristics, such as the most critical person or relationship. In contrast, macro-scale features reveal information about the network itself, such as the density of connections in the network. Mesoscale features existing in between provide distinct information, such as communities inside the network. Some examples of social networks include social media networks, business networks, and information circulation networks.
Several studies have been conducted using network analysis to tackle systems thinking-related topics. For instance, Dowd et al. (2018) proposed a new comprehensive decision-making framework for the maintenance, upgrading, and modernization of aging transportation infrastructure that includes highways, railways, bridges, and navigable waterways. This framework has three stages: (1) employ a systemic thinking approach to identify the impact factors, (2) complex network analysis to assess the criticality of the location of each component within the system, (3) apply a learning method to eliminate judgment and reduce the subjective disadvantages of learning algorithms currently available when using real data.
The results showed that this framework provides decision-makers with an index number representing the need for maintenance and modernization of each project and a hierarchical list in terms of essentiality. Tang et al. (2019) addressed the cascading effects in urban areas of earthquakes. The network analysis approach disclosed interrelated secondary events to generate ideas for designed strategies to mitigate disaster from systems thinking perspectives.
Network analysis is also applied in disaster risk management. In order to comprehend public security risks with associations, Tang and Lai (2019) suggested an approach to measure, analyze the public security risks, and provide an overview of effective mitigation strategies using social network analysis. The results showed that the framework added new value to the traditional risk management paradigm by offering support to urban managers developing comprehensive risk alleviation systems, hence, reducing risk interactions and their spread. Shiue et al. (2010) examined the social ties and perceived risk by using anonymity, offline activities, understanding quality, and media richness. This study showed that perceived risk and social ties were critical segments of social loafing. Social ties and perceived risk are hypothesized to affect social loafing within the online community.
Priven and Sacks (2015) adopted social network analysis to measure and analyze the communication levels between subcontractors. Pizzol and Scotti (2013) employed a systems thinking approach to identify marginal suppliers with geographical market delimitations where geographical markets are determined using network-based clustering. The results of this study showed that the proposed method could be applied to different products for which the trade network data are accessible, and it helps to provide a detailed analysis of technological constraints.
Health information available through social media has the potential to affect public health, especially when the population grows old. However, there is a lack of information about age-associated diseases on popular social media. Robillard et al. (2013) evaluated social media usage to share information about dementia, what sources of information about dementia were shared, and which dementia subjects commanded the conversation. The results showed that the dementia research community might benefit from the Twitter medium.
Health-related communities use social media frequently. Liu et al. (2016) investigated diabetes-related individuals on Twitter by depicting the frequency and timing of diabetes-related tweets, the geography of tweets, and the type of members over a two-year test of 10% of all tweets. The results showed that Twitter was becoming a space for online conversations about diabetes. Malik et al. (2014) evaluated the human advice-seeking behavior of primary health care (PHC) physicians. The study showed that SNA supported provider needs and presenting informal social interaction with physicians helped them use their professional linkages to seek advice since there was a lack of competent supervisory staff and a lack of improving functional indicators rather than clinical guidance. Blanchet et al. (2014) examined the sustainability indicators in rehabilitation sectors, contemplating differences in the governance and shape of the existing in two rehabilitation sectors. Differences in the structure of social networks marked greater after two years.
Lastly, in the tourism field, William et al. (2015) explored the content and structure of online word-of-mouth (eWOM) and its effect at a tourism destination when a festival was staged using both social network and text analysis. The results showed that Twitter users trusted seemingly disinterested opinions from individuals outside their social network.
Twitter, standing as one of the most widely used social media, provides various data such as tweet and retweet statistics, demographics, relations to followers, hashtags for particular topics. Global information exchange favors Twitter to be a platform where users read and post messages known as 'tweets' connecting and interacting with various communities. Users share their daily life interests, post their opinions on timely subjects such as brands and places. This platform has been considered an important access point for information and data. For instance, Tsimonis and Dimitriadis (2013) evaluated why organizations created brand pages in social media, how they used them, and what approaches, policies, and procedures they followed based on collected data from Twitter and other social media platforms. The authors recognized different opportunities for organization managers recommending practices for powerful and effective social media handling. Using data collected from Twitter, Koo et al. (2011) explored how an employee could utilize social communication technologies to suit his/her challenge characteristics because of the growing significance of social communication technologies within an organization. Additionally, the worker’s social relationships were tested in terms of moderate media use in job surroundings and the way this use encouraged the undertaking performance of selecting telephone, video conferencing, email, immediate messaging, and blog was tested. The use of social technology resulted in positive challenge performance. Rui et al. (2012) conducted a study where they examined whether and how Twitter WOM affected film sales through a dynamic panel fact model using a public data and machine learning algorithm. The results of this study showed that the effect of WOM on product sales from Twitter was higher when users had more followers compared with users with fewer followers. Moreover, tracking and analyzing people’s sentiments and intentions on Twitter showed that negative WOM decreased product sales, whereas positive WOM increased them. Kontopoulos et al. (2013) proposed an authentic ontology-based approach towards a more efficient and powerful sentiment analysis of Twitter posts. The purpose of this approach was to analyze the received sentiment from each distinct motion in a post instead of characterizing the posts by a sentiment score that was used by machine learning-based classifiers. The Twitter analysis is also used in the political domain to provide important information about individuals` thoughts, beliefs, and expectations. Golbeck and Hansen (2014) proposed a new approach to estimate and compute the political preferences among an organization’s Twitter followers. Twitter analysis is also used to distinguish agriculture area. Bastos et al. (2018) used the generic versus specialized information shared on Twitter network topology to identify communities and associations matching the agricultural area. The results of this investigation showed that decentralization increases when the information shared was generic and that the network adopts centralization formations as conversations became more specialized. Lastly, Aharony (2010) conducted a study to explore the use of Twitter in academic and public libraries to comprehend microblogging patterns. The study showed that there were some contrasts between academic and public libraries, including the number of tweets, linguistic differences, and content.
Methodology and data collection
The research goal based on the research gap
As one of the most significant scientific sources globally, the Scopus database was used to systematically explore the literature. This systematic review's main goal was to find the articles that used Twitter analysis to assess levels of systems thinking capabilities of individuals or reveal unique network patterns observed for systems thinkers. As mentioned in the Table 4, a general search thread to find the relevant research articles in related subject areas, keywords, and journals published in the English language was conducted. Consequently, 235 articles were listed by the Scopus database. A careful review of these articles revealed that the literature lacks studies investigating the level of systems thinking of individuals in connection to their Twitter network profiles. Motivated by the research problem and the identified literature gap, a focused and critical case study was designed. The worldwide outbreak of the COVID-19 pandemic, a vital and complex problem, emerged as a potential use case to investigate the relationship between systems thinkers’ Twitter responses to complex problems and their systems thinking capabilities. The assessment of the systems thinking skills of trusted Twitter accounts related to COVID-19 tweets was based on the influence these Twitter accounts have on their worldwide followers and societies. The procedure of identifying these 55 Twitter accounts and corresponding data collections is introduced in the next sections. The research framework is presented in Fig. 1.
Sample of population
In December 2019, the coronavirus (COVID-19) pandemic started to spread around the globe originally from China. In March 2020, the virus influenced almost most of the countries on every continent worldwide; consequently, the World Health Organization (WHO) declared the COVID-19 outbreak a worldwide pandemic. This virus is considered a danger to human health, resulting in worldwide fear and governmental quarantines. The side effect of this virus was somehow as dangerous as the virus itself for society. “As reported by the BBC, false information ranging from suggested medical care to conspiracy theories have been seen widely around the web, shared, and reposted by thousands of people on social media” (Wylde 2020); increased the number of casualties and pandemic fear. As a result of these threats, some credible media such as Forbes (Brown 2020), Fortune (Moore 2020), and Bustle (Wylde 2020) made up-to-date lists of trusted Twitter accounts of public health officials, researchers, epidemiologists, virus experts, family doctors, among others to assist in spreading accurate information, news, and recommendations to the masses to enhance society knowledge regarding COVID-19. The three trusted lists (i.e., Forbes, Fortune, and Bustle) were used in this research to identify credible organizations and individuals on Twitter as a sample of the current research population. A list of 55 Twitter accounts consists of 12 organizations such as WHO, GHS, and 43 individuals, have been selected for data collection. Table 5 presents the 55 Twitter accounts with information related to their account name, type of account, location, and the bio description on the corresponding Twitter account.
Data collection procedure and twitter features dataset
To collect the data, we have applied a team developer access to Twitter through Application Programming Interface (API). After a few rounds of review, Twitter grants us team access to collect data. Twitter, by default, has some restrictions for tweets. The first restriction is it gives a maximum of 5000 tweets for each individual account. The second restriction is that we can access the information of 15 followers of each account every 15 min, which makes the follower analysis time-consuming and difficult. These two restrictions were the main limitations of data collection for the current research. R version 4.0 and “rtweet” library have been used to collect the data from Twitter. To start tweet collection, the “create_token” function using four API credentials from Twitter, including consumer key, secret consumer key, access token, and access secret, have been utilized, followed by the usage of “get_timeline” function to extract tweets of 55 identified Twitter accounts. First, the potentially important information from each Twitter account was extracted, such as the name of the account, id, screen name, location, description, follower counts, friends count, listed count, favorites count, status counts, and some other information. Then, all the tweets for each account were categorized into (1) organic tweets, (2) replies, and (3) retweets. The ratios for each class calculated for every Twitter account, as shown in Table 6.
Follower analysis and features’ extraction
R version 4.0 and rtweet, tidyverse libraries were used to collect and analyze the followers’ data of 55 identified Twitter accounts. The follower analysis was performed based on the modified code from Bellman (2018). Due to the main restriction of Twitter regarding followers’ data, only a sample of 45 followers is selected to perform the follower network analysis. To collect data from a sample of 45 followers of each account, 45 min needed, and a total of 55 × 45 min needed to collect the followers' network information for all identified Twitter accounts. After collecting the followers’ data for each Twitter account, we extracted some information regarding the number of followers, following, and friends of each account. Then followers’ networks were constructed to show the relationship between the identified Twitter accounts and their corresponding followers. Additionally, the number of nodes and edges in each followers’ network extracted after network construction. First, we constructed a general connection between Twitter users and their followers. Figure 2 depicts the followers’ network of the “University of Washington Virology (UWV)” and “World Health Organization (WHO)” Twitter accounts; the nodes represent Twitter followers, and edges represent the connection between Twitter accounts.
The two following networks are a one-step neighborhood network meaning only two layers of followers were considered. For example, the University of Washington Virology’s followers and followers of those followers were collected with the condition that followers of followers that were not connected with first-level followers of UWV were removed from the network.
Nodes were defined as users, and edges connected two users based on the “following relationship”. The intensity of a node was quantified by the centrality of that node.
VOSviewer software was used to cluster the network. VOSviewer’s clustering algorithm has two hyperparameters, such as the minimum size of a cluster, a threshold for the minimum size of a cluster, and a resolution parameter that helps handle the number of clusters. VOSviewer uses a local moving algorithm introduced by Van Eck et al. (2013) to solve this optimization problem (determining the distinct clusters). Figure 2, top and bottom, shows the UWV and WHO followers network consisting of 6 and 5 distinct clusters, respectively. Four centrality measures, namely, Betweenness centrality, Closeness centrality, Degree centrality, and Eigen Centrality, were chosen for followers’ network analysis, and the summary statistics for each measure (i.e., mean centrality, SD centrality, median centrality, min centrality, and max centrality) were generated as network features. The node-level metrics, such as “this account follows how many accounts,” “number of this account’s friends,” “number of nodes in the network of this account,” and “number of edges in the network of this account,” were added to the features of followers’ network.
Result and discussion
Twitter network construction
The Twitter network is constructed, making use of 104 features extracted from 55 Twitter accounts. These features are summarized in Table 7. The first 12 features belong to Twitter account metrics. Next 40 features reflect some statistics related to all tweets, (1) favorite count, (2) retweet count, (3) quoted_favorite_count, (4) quoted_retweet_count, (5) quoted_followers_count, (6) quoted_friends_count, (7) quoted_statuses_count, (8) retweet_favorite_count, (9) retweet_retweet_count, (10) retweet_followers_count, (11) retweet_friends_count, and (12) retweet_statuses_count. The next 25 features are relevant to some statistics of organic tweets (no replies and retweets) such as favorite count, retweet count, quoted_favorite_count, quoted_retweet_count, quoted_friends_count, and quoted_statuses_count. The next 6 features are obtained from the sources that the identified Twitter accounts generate the content, followed by 11 features as the top ten most frequently used words and the frequency of COVID-19 and its synonyms in the tweets of each Twitter account. Finally, ten features are related to the sentimental scores of the tweets associated with each Twitter account.
Normalized Gower distance is used to quantify pairwise relationships among 55 accounts. Distance values are converted to a similarity matrix to construct the weighted network. Running a community structure finding (CSF) algorithm using the weighted network results in low modularity score (i.e. 0.072) detecting two clusters. CSF based on the modularity has resolution problem that it fails to find clusters at a finer resolution. In order to find clusters that result in better modularity of the network, the weighted network is binarized, retaining relationships stronger than 1.1 times the mean. Then, the community structure finding algorithm by Blondel et al. (2008) is applied to find similar Twitter accounts in the network. The algorithm detects three groups visualized in Fig. 3. These groups are further investigated to distinguish systems thinkers of various levels. It is critical to develop an objective strategy to map systems thinking qualities to some network measures. We made use of follower network features, including centrality scores and network-level metrics for mapping systems thinking dimensions to network measures.
Follower networks of 55 Twitter accounts, the followers’ features dataset are analyzed summarize statistics related to centrality and node-level metrics of each follower network (see Fig. 2) are extracted. Principle component analysis (PCA) is performed using follower data including betweenness, closeness, degree, eigen centralities, and node-level metrics. From five extracted features (that is, mean centrality, SD centrality, median centrality, min centrality, and max centrality) for betweenness centrality, one composite variable emerged. Additionally, one composite variable emerged for closeness and eigen centralities. Two composite variables emerged from degree centrality features. Table 1 indicates the summary of PCA for the centrality measures. To make all centrality composite variables the same scale, all the composite variables converted to percentage variables ranging from zero to 100 percent.
Moreover, the node-level metrics, such as out-degree (this account follows how many accounts), in-degree (number of this account’s friends), number of nodes in the network of this account, and number of edges in the network of this account were added to the five extracted composite centrality variables, as mentioned above. Finally, the five composite centrality variables and four node-level metrics were employed to interpret the followers’ characteristics of each of 55 Twitter accounts.
Systems thinking mapping
In this study, the Systems thinking (ST) skills instrument developed by Jaradat (2015) is utilized to assess the level of systemic thinking of individuals in the domain of complex systems problems. The ST skills instrument is developed based on the principles of systems science and theory in the area of complex systems. The ST skills instrument has seven dimensions that measure preferences toward seven systemic skills needed to engage effectively in complex system problems, as shown in Table 2. The seven ST dimensions are mapped with the extracted measures from follower analysis to find any existence of potential connections between follower analysis and level of ST skills of individuals.
The eigen centrality is an ideal ‘all-round’ social network analysis score, well suited for comprehending human social networks (Bhasin 2019; Bonacich 1972; Disney 2020; Franceschet 2018) that calculates the impact of a node (here, a Twitter account) using its degree and the degrees of its neighbors.
Eigen centrality has a hypothetical positive correlation with all seven ST skills dimensions. For example, the Twitter account that scores high eigen centrality is well-connected with its followers and has well-linked followers; the individual (who owns the Twitter account) might have a high level of Interaction and Autonomy skills based on the operational definitions in Table 2.
The account with a high score in the eigen centrality influences the entire network effectively and proficiently with a good amount of information and knowledge it has. Thus, the individual owner of Twitter might possess good systems thinking skills such as level of Hierarchical View and Complexity. Finally, the account with a high Eigen score is more responsive to explaining and solving unexpected and emergent social problems discussed in the Twitter network. Therefore, the Twitter account owner might have high systemic skills regarding the level of Change, Uncertainty, and Flexibility. Overall, the eigen centrality measure can be a good representative of the level of systemic thinking skills of the Twitter account owner.
The degree centrality score is similar to the eigen centrality score in the sense that measures consider the well-connectivity of a node in a network system (Bhasin 2019; Disney 2020; European Molecular Biology Laboratory 2020; Freeman 1978; Levorato 2014). Interestingly, two components emerged from our follower network PCA results, which can be related to in-degree and out-degree measures. The two degree-centrality measures are powerful in identifying the connected and popular people in the network system who probably have good amounts of knowledge and information; they can also quickly link with the wider network system. The in-degree measure can be related to levels of Hierarchical View, Complexity, and Interaction. out-degree can be related to levels of Autonomy, Change, Uncertainty, and Flexibility. Thus, the Degree Centrality might be a good indicator of an individual's level of systems thinking skills in a network of Twitter.
The closeness centrality assigns a score to a node according to its “closeness” to all other nodes (i.e., the shortest paths exist among all the nodes of a network) in the network system (Disney 2020; European Molecular Biology Laboratory 2020; Sabidussi 1966). closeness centrality can identify the best-placed people to affect the whole network system in the fastest time. This measure can distinguish the influencers and “broadcasters” who can exchange information and knowledge effectively and quickly. There are many similarities between the features of closeness centrality and the dimensions of ST skills. People with a high closeness score might have levels of Interaction, Change, Uncertainty, Hierarchical View, and Flexibility.
The betweenness centrality computes the frequency of a node poses on the shortest path among other nodes of the network (i.e., bridging between network’s nodes) (Disney 2020; European Molecular Biology Laboratory 2020; Levorato 2014; Freeman 1977; Borgatti 2005; Morselli and Roy 2008). This measure is useful to find how influential a person is on the flow around a network system. This measure is a good indicator of communication dynamics in a network. People with high betweenness scores tend to control information and hold authority in a network system (i.e., a gatekeeper in a network), and removing these individuals from the network system might disrupt communication, dynamics, and flow in the network system. According to the definition of the betweenness centrality measure and Table 2, there are hypothetical connections between the betweenness centrality measure and different dimensions of the ST instrument. For example, there is a connection between the level of interaction (that is, “interconnectedness in coordination and communication among multiple systems”) and the level of Hierarchical View (that is, “understanding system behavior at the whole versus part level”) and Betweenness centrality measure.
Based on foregoing discussion, the four measures of centrality can serve as good indicators for assessing the level of seven ST skills for Twitter accounts with respect to the analysis of their followers’ network, as depicted in Fig. 2.
Twitter features dataset (introduced in Sect. 4.A.) has been utilized to cluster 55 Twitter accounts. Since clustering is an unsupervised learning method that needs validation, the follower analysis results (introduced in Sect. 4.B.) are used to elucidate clusters. The emerged clusters of accounts shown in Fig. 3 are enriched based on account metrics pertaining to follower analysis. Table 3 shows the mean and SD of the Twitter analysis metric for each of the three clusters. Table 8 presents the details for all Twitter accounts grouped into three clusters. Tables 9 and 10 indicate the results of between-group ANOVA and Tukey HSD t-tests performed as post-hoc multiple-group comparison tests.
Interestingly, Twitter accounts in cluster 3 have relatively higher follower network centrality measures than the other two clusters. Additionally, Twitter accounts in cluster 1 have relatively lower follower network centrality metrics than those in cluster 2. According to mapping the relationships between ST dimensions and Twitter accounts' centrality measures, we infer high scores in centrality measures of follower analysis associated with high ST capability of individuals. As a result, we conclude since Twitter accounts in cluster 3 have relatively higher centrality measures, the individuals corresponding to these Twitter accounts might have higher ST capability skills than others. Consequently, we call Twitter accounts in cluster 3 holistic thinker cluster. Similarly, the Twitter accounts in cluster 1 are reductionist thinkers due to low scores pertaining to their follower network analysis's centrality scores. Since Twitter accounts in cluster 2 have centrality scores almost between clusters 1 and 3, they are called middle thinker Twitter accounts. Results indicate the follower network analysis might be able to quantify systemic thinking capabilities of individuals based on their tweets in the case of the COVID-19 pandemic.
Twitter is a crucial medium in circulating information, and unfortunately, sometimes erroneous or inadequate messages are propagated in a network by users. For example, in the case of the worldwide outbreak of the COVID-19 pandemic, we observed many unsupported claims about the origin of the virus, virus transmission, possible prevention, and likely treatment. In this confusing situation, knowing reliable sources and the level of holistic systems thinking of these sources greatly assist the public in making correct decisions. Thus, COVID-19 stands as a potential use case to investigate the relationship between COVID-19 experts’ Twitter network and their systems thinking capabilities.
The interdisciplinary literature review revealed that there was a literature gap connecting COVID-19 experts’ system thinking level and their Twitter network profiles. Therefore, we identified 55 trusted expert Twitter accounts based on the lists of Forbes, Fortune, and Bustle.
We constructed the Twitter network of these experts based on features extracted from their Twitter accounts, including organic metrics (e.g., account favorite count, list count, followers count, etc.), tweets’ measures (e.g., favorite count, retweet count, reply count, etc.), sentimental analysis, source of tweets as well as their tweets like most frequent words and their frequencies and frequency of using COVID-19 synonyms. Clustering the network based on tweets and Twitter features identified three distinct groups of experts emerged. The follower networks of experts were generated to elucidate clusters. The system thinking dimensions were mapped to follower network measures such as betweenness, closeness, degree, eigen centralities, and node-level metrics. By comparing the follower network measures of 55 experts, we inferred that three identified clusters had meaningful differences in centrality scores and node-level metrics. The cluster with a higher, medium, lower score can be classified as Holistic thinkers, Middle thinkers, and Reductionist thinkers, respectively.
Briefly, this research tested that the capabilities of individuals as system thinkers can reveal unique network patterns and distinct communities associated with the level of system thinking of COVID-19 experts.
Twitter restrictions related to the search of the past tweets and access to followers’ information of each Twitter account was one of the limitations of the study. Another limitation was the validity of mapping systems thinking dimensions to centrality measures of follower network analysis; further investigation with other samples of the population is needed to validate the current mapping. Future studies will be directed into investigating the relationship between Twitter users’ systems thinking and their sentiment analysis. Analyzing the organic tweets of expert Twitter accounts related to COVID-19 using text mining and natural language processing to better understand the impact of tweets on followers and spread knowledge and information within the Twitter network. The new research finding can be compared and combined with the current research results to analyze these studies' validity and consistency.
The results of the current study have some practical implications as follows:
Using follower analysis and Twitter analysis to understand systems thinking of people, especially the influencers and celebrities, and how their role is important to spread the true news and knowledge to the community.
How systems thinking is related to having a better network of followers promotes the more efficient and effective transformation of information and knowledge to the community.
As an individual’s level of systems thinking skills can be enhanced, the social media activity of the individuals can be improved. This is very important since the research shows there is a necessity to create a safe and healthy virtual environment for everybody, so everyone can express their opinions and beliefs in the direction of the community’s values.
Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.
Aharony N (2010) Twitter use in libraries: an exploratory analysis. J Web Librariansh 4(4):333–350
Bastos M, Piccardi C, Levy M, McRoberts N, Lubell M (2018) Core-periphery or decentralized? Topological shifts of specialized information on twitter. Soc Netw 52:282–293
Bellman B (2018) Twitter data and network analysis with R. RPubs by RStudio. https://rpubs.com/ben_bellman/rtweet_tidygraph
Bhasin J (2019) Graph analytics—introduction and concepts of centrality. Towards data science. https://towardsdatascience.com/graph-analytics-introduction-and-concepts-of-centrality-8f5543b55de3
Blanchet K, Palmer J, Palanchowke R, Boggs D, Jama A, Girois S (2014) Advancing the application of systems thinking in health: analyzing the contextual and social network factors influencing the use of sustainability indicators in a health system—a comparative study in Nepal and Somaliland. Heal Res Policy Syst 12(1):1
Blondel VD et al (2008) Fast unfolding of communities in large networks. J Stat Mech 2008:P10008
Boardman J, Sauser B (2006) System of systems-the meaning. In: 2006 IEEE/SMC international conference on system of systems engineering, IEEE, 6-pp
Bonacich P (1972) Factoring and weighting approaches to status scores and clique identification. J Math Sociol 2(1):113–120
Borgatti SP (2005) Centrality and network flow. Soc Netw 27(1):55
Brown A (2020) Coronavirus: the most essential people to follow on twitter during the COVID-19 outbreak. https://www.forbes.com/sites/abrambrown/2020/03/14/coronavirus-the-most-essential-people-on-twitter-to-follow-during-the-covid-19-outbreak/#1969611375f3
Clegg CW (2000) Sociotechnical principles for system design. Appl Ergon 31(5):463–477
DeLaurentis D (2005) Understanding transportation as a system-of-systems design problem. In: 43rd AIAA aerospace sciences meeting and exhibit, p 123
Disney A (2020) Social network analysis 101: centrality measures explained. Cambridge intelligence. https://cambridge-intelligence.com/keylines-faqs-social-network-analysis/#:~:text=Definition%3A%20Betweenness%20centrality%20measures%20the,shortest%20path%20between%20other%20nodes.&text=It%20does%20this%20by%20identifying,the%20flow%20around%20a%20system
Dowd Z, Franz AY, Wasek JS (2018) A decision-making framework for maintenance and modernization of transportation infrastructure. IEEE Trans Eng Manag 67:42
European Molecular Biology Laboratory (2020) Centrality analysis. https://www.ebi.ac.uk/training/online/course/network-analysis-protein-interaction-data-introduction/building-and-analysing-ppins-0
Franceschet M (2018) Network science. https://www.sci.unich.it/~francesc/teaching/network/
Frank M (2006) Knowledge, abilities, cognitive characteristics and behavioral competences of engineers with high capacity for engineering systems thinking (CEST). Syst Eng 9(2):91–103
Freeman LC (1977) A set of measures of centrality based on betweenness. Sociometry 40:35–41
Freeman LC (1978) Centrality in social networks conceptual clarification. Soc Netw 1(3):215–239
Golbeck J, Hansen D (2014) A method for computing political preference among Twitter followers. Social Net 36:177–184
Gorod A, Sauser B, Boardman J (2008) System-of-systems engineering management: a review of modern history and a path forward. IEEE Syst J 2(4):484–499
Hossain NUI, Dayarathna VL, Nagahi M, Jaradat R (2020) Systems thinking: a review and bibliometric analysis. Systems 8(3):23
Jaradat RM (2015) Complex system governance requires systems thinking-how to find systems thinkers. Int J Syst Eng 6(1–2):53–70
Jaradat RM, Keating CB, Bradley JM (2017) Individual capacity and organizational competency for systems thinking. IEEE Syst J12:2
Karam S, Nagahi M, Dayarathna VL, Ma J, Jaradat R, Hamilton M (2020) Integrating systems thinking skills with multi-criteria decision-making technology to recruit employee candidates. Expert Syst Appl 160:113585
Keating CB (2008) Emergence in system of systems. In: System of systems engineering: innovations for the 21st century, pp169–190
Kontopoulos E et al (2013) Ontology-based sentiment analysis of twitter posts. Expert Syst Appl 40(10):4065–4074
Koo C, Wati Y, Jason J (2011) Examination of how social aspects moderate the relationship between task characteristics and usage of social commination technologies (SCTs) in organizations. Int J Inf Manage 31(5):445–459
Levorato V (2014) Group measures and modeling for social networks. J Complex Syst
Liu Y, Mei Q, Hanauer DA, Zheng K, Lee JM (2016) Use of social media in the diabetes community: an exploratory analysis of diabetes-related tweets. JMIR Diabetes 1(2):e4
Malik AU, Wills CD, Hamid S, Ulikpan A, Hill PS (2014) Advancing the application of systems thinking in health: advice seeking behavior among primary health care physicians in Pakistan. Health Res Policy Syst 12(1):43
Moore M (2020) The best Twitter accounts to follow for reliable information on the coronavirus outbreak. https://fortune.com/2020/03/14/coronavirus-updates-twitter-accounts-covid-19-news/
Morselli C, Roy J (2008) Brokerage qualifications in ringing operations. Criminology 46(1):71–98
Nagahi M, Jaradat R, Goerger SR, Hamilton M, Buchanan RK, Abutabenjeh S, Ma J (2020) The impact of practitioners’ personality traits on their level of systems-thinking skills preferences. Eng Manag J 3:1–18
Nagahi M, Jaradat R, Amrani SE, Hamilton M, Goerger SR (2020b) Holistic and reductionist thinker: a comparison study based on individuals’ skillset and personality types. Int J Syst Syst Eng 10(4):337–366
Nagahi M, Hossain NUI, Jaradat R, Dayarathna V, Keating C, Goerger S, Hamilton M (2020c) Classification of individual managers' systems thinking skills based on different organizational ownership structures. In: Systems research and behavioral science
Pizzol M, Scotti M (2013) Identifying marginal supplying countries of wood products via trade network analysis. Int J Life Cycle Assess 22(7):1146–1158
Priven V, Sacks R (2015) Effects of the last planner system on social networking among construction trade crews. J Constr Eng Manag 141(6):04015006
Robillard JM, Johnson TW, Hennessey C, Lynn Beattie B, Illes J (2013) Aging 20: health information about dementia on twitter. PLoS One 8(7):e69861
Rui H, Liu Y, Whinston A (2012) Whose and what chatter matters? The effect of tweets on movie sales. Decision Support Syst 55(4):1
Sabidussi G (1966) The centrality index of a graph. Psychometrika 31
Shiue Y-C, Chiu C-M, Chang C-C (2010) Exploring and mitigating social loafing in online communities. Comput Hum Behav 26(4):768–777
Tang P, Lai S (2019) A framework for managing public security risks with complex interactions in cities and its application evidenced from Shenzhen City in China. Cities 95:102390
Tang P, Xia Qi, Wang Y (2019) Addressing cascading effects of earthquakes in urban areas from network perspective to improve disaster mitigation. Int Disaster Risk Reduct 35(2019):101065
Tsimonis G, Dimitriadis S (2013) Brand strategies in social media. In: Marketing intelligence and planning
Ullah Ibne Hossain N et al (2020) The effect of an individual’s education level on their systems skills in the system of systems domain. J Manag Anal 7(4):510–531
Van Eck NJ, Waltman L (2013) VOSviewer manual, vol 1. Univeristeit Leiden, Leiden, pp 1–53
Williams NL, Inversini A, Buhalis D, Ferdinand N (2015) Community crosstalk: an exploratory analysis of destination and festival eWOM on Twitter. J Mark Manag 31(9–10):1113–1140
Wylde K (2020) March 18. Expert Twitter accounts for coronavirus and COVID-19 updates. https://www.bustle.com/p/expert-twitter-accounts-for-coronavirus-covid-19-updates-22419348
Ethics approval and consent to participate
Consent for publication
There are no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Pirim, H., Nagahi, M., Larif, O. et al. Integrated twitter analysis to distinguish systems thinkers at various levels: a case study of COVID-19. Appl Netw Sci 8, 12 (2023). https://doi.org/10.1007/s41109-022-00520-9
- Systems thinking
- Systems skills
- Social networks
- Twitter analysis
- Follower network
- Network clustering